1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_da_format.h"
17 #include "xfs_da_btree.h"
18 #include "xfs_inode.h"
19 #include "xfs_dir2.h"
20 #include "xfs_ialloc.h"
21 #include "xfs_alloc.h"
22 #include "xfs_rtalloc.h"
23 #include "xfs_bmap.h"
24 #include "xfs_trans.h"
25 #include "xfs_trans_priv.h"
26 #include "xfs_log.h"
27 #include "xfs_error.h"
28 #include "xfs_quota.h"
29 #include "xfs_fsops.h"
30 #include "xfs_trace.h"
31 #include "xfs_icache.h"
32 #include "xfs_sysfs.h"
33 #include "xfs_rmap_btree.h"
34 #include "xfs_refcount_btree.h"
35 #include "xfs_reflink.h"
36 #include "xfs_extent_busy.h"
37 
38 
39 static DEFINE_MUTEX(xfs_uuid_table_mutex);
40 static int xfs_uuid_table_size;
41 static uuid_t *xfs_uuid_table;
42 
43 void
xfs_uuid_table_free(void)44 xfs_uuid_table_free(void)
45 {
46 	if (xfs_uuid_table_size == 0)
47 		return;
48 	kmem_free(xfs_uuid_table);
49 	xfs_uuid_table = NULL;
50 	xfs_uuid_table_size = 0;
51 }
52 
53 /*
54  * See if the UUID is unique among mounted XFS filesystems.
55  * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
56  */
57 STATIC int
xfs_uuid_mount(struct xfs_mount * mp)58 xfs_uuid_mount(
59 	struct xfs_mount	*mp)
60 {
61 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
62 	int			hole, i;
63 
64 	/* Publish UUID in struct super_block */
65 	uuid_copy(&mp->m_super->s_uuid, uuid);
66 
67 	if (mp->m_flags & XFS_MOUNT_NOUUID)
68 		return 0;
69 
70 	if (uuid_is_null(uuid)) {
71 		xfs_warn(mp, "Filesystem has null UUID - can't mount");
72 		return -EINVAL;
73 	}
74 
75 	mutex_lock(&xfs_uuid_table_mutex);
76 	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
77 		if (uuid_is_null(&xfs_uuid_table[i])) {
78 			hole = i;
79 			continue;
80 		}
81 		if (uuid_equal(uuid, &xfs_uuid_table[i]))
82 			goto out_duplicate;
83 	}
84 
85 	if (hole < 0) {
86 		xfs_uuid_table = kmem_realloc(xfs_uuid_table,
87 			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
88 			KM_SLEEP);
89 		hole = xfs_uuid_table_size++;
90 	}
91 	xfs_uuid_table[hole] = *uuid;
92 	mutex_unlock(&xfs_uuid_table_mutex);
93 
94 	return 0;
95 
96  out_duplicate:
97 	mutex_unlock(&xfs_uuid_table_mutex);
98 	xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
99 	return -EINVAL;
100 }
101 
102 STATIC void
xfs_uuid_unmount(struct xfs_mount * mp)103 xfs_uuid_unmount(
104 	struct xfs_mount	*mp)
105 {
106 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
107 	int			i;
108 
109 	if (mp->m_flags & XFS_MOUNT_NOUUID)
110 		return;
111 
112 	mutex_lock(&xfs_uuid_table_mutex);
113 	for (i = 0; i < xfs_uuid_table_size; i++) {
114 		if (uuid_is_null(&xfs_uuid_table[i]))
115 			continue;
116 		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
117 			continue;
118 		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
119 		break;
120 	}
121 	ASSERT(i < xfs_uuid_table_size);
122 	mutex_unlock(&xfs_uuid_table_mutex);
123 }
124 
125 
126 STATIC void
__xfs_free_perag(struct rcu_head * head)127 __xfs_free_perag(
128 	struct rcu_head	*head)
129 {
130 	struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
131 
132 	ASSERT(atomic_read(&pag->pag_ref) == 0);
133 	kmem_free(pag);
134 }
135 
136 /*
137  * Free up the per-ag resources associated with the mount structure.
138  */
139 STATIC void
xfs_free_perag(xfs_mount_t * mp)140 xfs_free_perag(
141 	xfs_mount_t	*mp)
142 {
143 	xfs_agnumber_t	agno;
144 	struct xfs_perag *pag;
145 
146 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
147 		spin_lock(&mp->m_perag_lock);
148 		pag = radix_tree_delete(&mp->m_perag_tree, agno);
149 		spin_unlock(&mp->m_perag_lock);
150 		ASSERT(pag);
151 		ASSERT(atomic_read(&pag->pag_ref) == 0);
152 		xfs_buf_hash_destroy(pag);
153 		mutex_destroy(&pag->pag_ici_reclaim_lock);
154 		call_rcu(&pag->rcu_head, __xfs_free_perag);
155 	}
156 }
157 
158 /*
159  * Check size of device based on the (data/realtime) block count.
160  * Note: this check is used by the growfs code as well as mount.
161  */
162 int
xfs_sb_validate_fsb_count(xfs_sb_t * sbp,uint64_t nblocks)163 xfs_sb_validate_fsb_count(
164 	xfs_sb_t	*sbp,
165 	uint64_t	nblocks)
166 {
167 	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
168 	ASSERT(sbp->sb_blocklog >= BBSHIFT);
169 
170 	/* Limited by ULONG_MAX of page cache index */
171 	if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
172 		return -EFBIG;
173 	return 0;
174 }
175 
176 int
xfs_initialize_perag(xfs_mount_t * mp,xfs_agnumber_t agcount,xfs_agnumber_t * maxagi)177 xfs_initialize_perag(
178 	xfs_mount_t	*mp,
179 	xfs_agnumber_t	agcount,
180 	xfs_agnumber_t	*maxagi)
181 {
182 	xfs_agnumber_t	index;
183 	xfs_agnumber_t	first_initialised = NULLAGNUMBER;
184 	xfs_perag_t	*pag;
185 	int		error = -ENOMEM;
186 
187 	/*
188 	 * Walk the current per-ag tree so we don't try to initialise AGs
189 	 * that already exist (growfs case). Allocate and insert all the
190 	 * AGs we don't find ready for initialisation.
191 	 */
192 	for (index = 0; index < agcount; index++) {
193 		pag = xfs_perag_get(mp, index);
194 		if (pag) {
195 			xfs_perag_put(pag);
196 			continue;
197 		}
198 
199 		pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
200 		if (!pag)
201 			goto out_unwind_new_pags;
202 		pag->pag_agno = index;
203 		pag->pag_mount = mp;
204 		spin_lock_init(&pag->pag_ici_lock);
205 		mutex_init(&pag->pag_ici_reclaim_lock);
206 		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
207 		if (xfs_buf_hash_init(pag))
208 			goto out_free_pag;
209 		init_waitqueue_head(&pag->pagb_wait);
210 		spin_lock_init(&pag->pagb_lock);
211 		pag->pagb_count = 0;
212 		pag->pagb_tree = RB_ROOT;
213 
214 		if (radix_tree_preload(GFP_NOFS))
215 			goto out_hash_destroy;
216 
217 		spin_lock(&mp->m_perag_lock);
218 		if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
219 			BUG();
220 			spin_unlock(&mp->m_perag_lock);
221 			radix_tree_preload_end();
222 			error = -EEXIST;
223 			goto out_hash_destroy;
224 		}
225 		spin_unlock(&mp->m_perag_lock);
226 		radix_tree_preload_end();
227 		/* first new pag is fully initialized */
228 		if (first_initialised == NULLAGNUMBER)
229 			first_initialised = index;
230 	}
231 
232 	index = xfs_set_inode_alloc(mp, agcount);
233 
234 	if (maxagi)
235 		*maxagi = index;
236 
237 	mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
238 	return 0;
239 
240 out_hash_destroy:
241 	xfs_buf_hash_destroy(pag);
242 out_free_pag:
243 	mutex_destroy(&pag->pag_ici_reclaim_lock);
244 	kmem_free(pag);
245 out_unwind_new_pags:
246 	/* unwind any prior newly initialized pags */
247 	for (index = first_initialised; index < agcount; index++) {
248 		pag = radix_tree_delete(&mp->m_perag_tree, index);
249 		if (!pag)
250 			break;
251 		xfs_buf_hash_destroy(pag);
252 		mutex_destroy(&pag->pag_ici_reclaim_lock);
253 		kmem_free(pag);
254 	}
255 	return error;
256 }
257 
258 /*
259  * xfs_readsb
260  *
261  * Does the initial read of the superblock.
262  */
263 int
xfs_readsb(struct xfs_mount * mp,int flags)264 xfs_readsb(
265 	struct xfs_mount *mp,
266 	int		flags)
267 {
268 	unsigned int	sector_size;
269 	struct xfs_buf	*bp;
270 	struct xfs_sb	*sbp = &mp->m_sb;
271 	int		error;
272 	int		loud = !(flags & XFS_MFSI_QUIET);
273 	const struct xfs_buf_ops *buf_ops;
274 
275 	ASSERT(mp->m_sb_bp == NULL);
276 	ASSERT(mp->m_ddev_targp != NULL);
277 
278 	/*
279 	 * For the initial read, we must guess at the sector
280 	 * size based on the block device.  It's enough to
281 	 * get the sb_sectsize out of the superblock and
282 	 * then reread with the proper length.
283 	 * We don't verify it yet, because it may not be complete.
284 	 */
285 	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
286 	buf_ops = NULL;
287 
288 	/*
289 	 * Allocate a (locked) buffer to hold the superblock. This will be kept
290 	 * around at all times to optimize access to the superblock. Therefore,
291 	 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
292 	 * elevated.
293 	 */
294 reread:
295 	error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
296 				      BTOBB(sector_size), XBF_NO_IOACCT, &bp,
297 				      buf_ops);
298 	if (error) {
299 		if (loud)
300 			xfs_warn(mp, "SB validate failed with error %d.", error);
301 		/* bad CRC means corrupted metadata */
302 		if (error == -EFSBADCRC)
303 			error = -EFSCORRUPTED;
304 		return error;
305 	}
306 
307 	/*
308 	 * Initialize the mount structure from the superblock.
309 	 */
310 	xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
311 
312 	/*
313 	 * If we haven't validated the superblock, do so now before we try
314 	 * to check the sector size and reread the superblock appropriately.
315 	 */
316 	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
317 		if (loud)
318 			xfs_warn(mp, "Invalid superblock magic number");
319 		error = -EINVAL;
320 		goto release_buf;
321 	}
322 
323 	/*
324 	 * We must be able to do sector-sized and sector-aligned IO.
325 	 */
326 	if (sector_size > sbp->sb_sectsize) {
327 		if (loud)
328 			xfs_warn(mp, "device supports %u byte sectors (not %u)",
329 				sector_size, sbp->sb_sectsize);
330 		error = -ENOSYS;
331 		goto release_buf;
332 	}
333 
334 	if (buf_ops == NULL) {
335 		/*
336 		 * Re-read the superblock so the buffer is correctly sized,
337 		 * and properly verified.
338 		 */
339 		xfs_buf_relse(bp);
340 		sector_size = sbp->sb_sectsize;
341 		buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
342 		goto reread;
343 	}
344 
345 	xfs_reinit_percpu_counters(mp);
346 
347 	/* no need to be quiet anymore, so reset the buf ops */
348 	bp->b_ops = &xfs_sb_buf_ops;
349 
350 	mp->m_sb_bp = bp;
351 	xfs_buf_unlock(bp);
352 	return 0;
353 
354 release_buf:
355 	xfs_buf_relse(bp);
356 	return error;
357 }
358 
359 /*
360  * Update alignment values based on mount options and sb values
361  */
362 STATIC int
xfs_update_alignment(xfs_mount_t * mp)363 xfs_update_alignment(xfs_mount_t *mp)
364 {
365 	xfs_sb_t	*sbp = &(mp->m_sb);
366 
367 	if (mp->m_dalign) {
368 		/*
369 		 * If stripe unit and stripe width are not multiples
370 		 * of the fs blocksize turn off alignment.
371 		 */
372 		if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
373 		    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
374 			xfs_warn(mp,
375 		"alignment check failed: sunit/swidth vs. blocksize(%d)",
376 				sbp->sb_blocksize);
377 			return -EINVAL;
378 		} else {
379 			/*
380 			 * Convert the stripe unit and width to FSBs.
381 			 */
382 			mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
383 			if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
384 				xfs_warn(mp,
385 			"alignment check failed: sunit/swidth vs. agsize(%d)",
386 					 sbp->sb_agblocks);
387 				return -EINVAL;
388 			} else if (mp->m_dalign) {
389 				mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
390 			} else {
391 				xfs_warn(mp,
392 			"alignment check failed: sunit(%d) less than bsize(%d)",
393 					 mp->m_dalign, sbp->sb_blocksize);
394 				return -EINVAL;
395 			}
396 		}
397 
398 		/*
399 		 * Update superblock with new values
400 		 * and log changes
401 		 */
402 		if (xfs_sb_version_hasdalign(sbp)) {
403 			if (sbp->sb_unit != mp->m_dalign) {
404 				sbp->sb_unit = mp->m_dalign;
405 				mp->m_update_sb = true;
406 			}
407 			if (sbp->sb_width != mp->m_swidth) {
408 				sbp->sb_width = mp->m_swidth;
409 				mp->m_update_sb = true;
410 			}
411 		} else {
412 			xfs_warn(mp,
413 	"cannot change alignment: superblock does not support data alignment");
414 			return -EINVAL;
415 		}
416 	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
417 		    xfs_sb_version_hasdalign(&mp->m_sb)) {
418 			mp->m_dalign = sbp->sb_unit;
419 			mp->m_swidth = sbp->sb_width;
420 	}
421 
422 	return 0;
423 }
424 
425 /*
426  * Set the maximum inode count for this filesystem
427  */
428 STATIC void
xfs_set_maxicount(xfs_mount_t * mp)429 xfs_set_maxicount(xfs_mount_t *mp)
430 {
431 	xfs_sb_t	*sbp = &(mp->m_sb);
432 	uint64_t	icount;
433 
434 	if (sbp->sb_imax_pct) {
435 		/*
436 		 * Make sure the maximum inode count is a multiple
437 		 * of the units we allocate inodes in.
438 		 */
439 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
440 		do_div(icount, 100);
441 		do_div(icount, mp->m_ialloc_blks);
442 		mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
443 				   sbp->sb_inopblog;
444 	} else {
445 		mp->m_maxicount = 0;
446 	}
447 }
448 
449 /*
450  * Set the default minimum read and write sizes unless
451  * already specified in a mount option.
452  * We use smaller I/O sizes when the file system
453  * is being used for NFS service (wsync mount option).
454  */
455 STATIC void
xfs_set_rw_sizes(xfs_mount_t * mp)456 xfs_set_rw_sizes(xfs_mount_t *mp)
457 {
458 	xfs_sb_t	*sbp = &(mp->m_sb);
459 	int		readio_log, writeio_log;
460 
461 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
462 		if (mp->m_flags & XFS_MOUNT_WSYNC) {
463 			readio_log = XFS_WSYNC_READIO_LOG;
464 			writeio_log = XFS_WSYNC_WRITEIO_LOG;
465 		} else {
466 			readio_log = XFS_READIO_LOG_LARGE;
467 			writeio_log = XFS_WRITEIO_LOG_LARGE;
468 		}
469 	} else {
470 		readio_log = mp->m_readio_log;
471 		writeio_log = mp->m_writeio_log;
472 	}
473 
474 	if (sbp->sb_blocklog > readio_log) {
475 		mp->m_readio_log = sbp->sb_blocklog;
476 	} else {
477 		mp->m_readio_log = readio_log;
478 	}
479 	mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
480 	if (sbp->sb_blocklog > writeio_log) {
481 		mp->m_writeio_log = sbp->sb_blocklog;
482 	} else {
483 		mp->m_writeio_log = writeio_log;
484 	}
485 	mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
486 }
487 
488 /*
489  * precalculate the low space thresholds for dynamic speculative preallocation.
490  */
491 void
xfs_set_low_space_thresholds(struct xfs_mount * mp)492 xfs_set_low_space_thresholds(
493 	struct xfs_mount	*mp)
494 {
495 	int i;
496 
497 	for (i = 0; i < XFS_LOWSP_MAX; i++) {
498 		uint64_t space = mp->m_sb.sb_dblocks;
499 
500 		do_div(space, 100);
501 		mp->m_low_space[i] = space * (i + 1);
502 	}
503 }
504 
505 
506 /*
507  * Set whether we're using inode alignment.
508  */
509 STATIC void
xfs_set_inoalignment(xfs_mount_t * mp)510 xfs_set_inoalignment(xfs_mount_t *mp)
511 {
512 	if (xfs_sb_version_hasalign(&mp->m_sb) &&
513 		mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp))
514 		mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
515 	else
516 		mp->m_inoalign_mask = 0;
517 	/*
518 	 * If we are using stripe alignment, check whether
519 	 * the stripe unit is a multiple of the inode alignment
520 	 */
521 	if (mp->m_dalign && mp->m_inoalign_mask &&
522 	    !(mp->m_dalign & mp->m_inoalign_mask))
523 		mp->m_sinoalign = mp->m_dalign;
524 	else
525 		mp->m_sinoalign = 0;
526 }
527 
528 /*
529  * Check that the data (and log if separate) is an ok size.
530  */
531 STATIC int
xfs_check_sizes(struct xfs_mount * mp)532 xfs_check_sizes(
533 	struct xfs_mount *mp)
534 {
535 	struct xfs_buf	*bp;
536 	xfs_daddr_t	d;
537 	int		error;
538 
539 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
540 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
541 		xfs_warn(mp, "filesystem size mismatch detected");
542 		return -EFBIG;
543 	}
544 	error = xfs_buf_read_uncached(mp->m_ddev_targp,
545 					d - XFS_FSS_TO_BB(mp, 1),
546 					XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
547 	if (error) {
548 		xfs_warn(mp, "last sector read failed");
549 		return error;
550 	}
551 	xfs_buf_relse(bp);
552 
553 	if (mp->m_logdev_targp == mp->m_ddev_targp)
554 		return 0;
555 
556 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
557 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
558 		xfs_warn(mp, "log size mismatch detected");
559 		return -EFBIG;
560 	}
561 	error = xfs_buf_read_uncached(mp->m_logdev_targp,
562 					d - XFS_FSB_TO_BB(mp, 1),
563 					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
564 	if (error) {
565 		xfs_warn(mp, "log device read failed");
566 		return error;
567 	}
568 	xfs_buf_relse(bp);
569 	return 0;
570 }
571 
572 /*
573  * Clear the quotaflags in memory and in the superblock.
574  */
575 int
xfs_mount_reset_sbqflags(struct xfs_mount * mp)576 xfs_mount_reset_sbqflags(
577 	struct xfs_mount	*mp)
578 {
579 	mp->m_qflags = 0;
580 
581 	/* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
582 	if (mp->m_sb.sb_qflags == 0)
583 		return 0;
584 	spin_lock(&mp->m_sb_lock);
585 	mp->m_sb.sb_qflags = 0;
586 	spin_unlock(&mp->m_sb_lock);
587 
588 	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
589 		return 0;
590 
591 	return xfs_sync_sb(mp, false);
592 }
593 
594 uint64_t
xfs_default_resblks(xfs_mount_t * mp)595 xfs_default_resblks(xfs_mount_t *mp)
596 {
597 	uint64_t resblks;
598 
599 	/*
600 	 * We default to 5% or 8192 fsbs of space reserved, whichever is
601 	 * smaller.  This is intended to cover concurrent allocation
602 	 * transactions when we initially hit enospc. These each require a 4
603 	 * block reservation. Hence by default we cover roughly 2000 concurrent
604 	 * allocation reservations.
605 	 */
606 	resblks = mp->m_sb.sb_dblocks;
607 	do_div(resblks, 20);
608 	resblks = min_t(uint64_t, resblks, 8192);
609 	return resblks;
610 }
611 
612 /* Ensure the summary counts are correct. */
613 STATIC int
xfs_check_summary_counts(struct xfs_mount * mp)614 xfs_check_summary_counts(
615 	struct xfs_mount	*mp)
616 {
617 	/*
618 	 * The AG0 superblock verifier rejects in-progress filesystems,
619 	 * so we should never see the flag set this far into mounting.
620 	 */
621 	if (mp->m_sb.sb_inprogress) {
622 		xfs_err(mp, "sb_inprogress set after log recovery??");
623 		WARN_ON(1);
624 		return -EFSCORRUPTED;
625 	}
626 
627 	/*
628 	 * Now the log is mounted, we know if it was an unclean shutdown or
629 	 * not. If it was, with the first phase of recovery has completed, we
630 	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
631 	 * but they are recovered transactionally in the second recovery phase
632 	 * later.
633 	 *
634 	 * If the log was clean when we mounted, we can check the summary
635 	 * counters.  If any of them are obviously incorrect, we can recompute
636 	 * them from the AGF headers in the next step.
637 	 */
638 	if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
639 	    (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
640 	     !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
641 	     mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
642 		mp->m_flags |= XFS_MOUNT_BAD_SUMMARY;
643 
644 	/*
645 	 * We can safely re-initialise incore superblock counters from the
646 	 * per-ag data. These may not be correct if the filesystem was not
647 	 * cleanly unmounted, so we waited for recovery to finish before doing
648 	 * this.
649 	 *
650 	 * If the filesystem was cleanly unmounted or the previous check did
651 	 * not flag anything weird, then we can trust the values in the
652 	 * superblock to be correct and we don't need to do anything here.
653 	 * Otherwise, recalculate the summary counters.
654 	 */
655 	if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) ||
656 	     XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) &&
657 	    !(mp->m_flags & XFS_MOUNT_BAD_SUMMARY))
658 		return 0;
659 
660 	return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
661 }
662 
663 /*
664  * This function does the following on an initial mount of a file system:
665  *	- reads the superblock from disk and init the mount struct
666  *	- if we're a 32-bit kernel, do a size check on the superblock
667  *		so we don't mount terabyte filesystems
668  *	- init mount struct realtime fields
669  *	- allocate inode hash table for fs
670  *	- init directory manager
671  *	- perform recovery and init the log manager
672  */
673 int
xfs_mountfs(struct xfs_mount * mp)674 xfs_mountfs(
675 	struct xfs_mount	*mp)
676 {
677 	struct xfs_sb		*sbp = &(mp->m_sb);
678 	struct xfs_inode	*rip;
679 	uint64_t		resblks;
680 	uint			quotamount = 0;
681 	uint			quotaflags = 0;
682 	int			error = 0;
683 
684 	xfs_sb_mount_common(mp, sbp);
685 
686 	/*
687 	 * Check for a mismatched features2 values.  Older kernels read & wrote
688 	 * into the wrong sb offset for sb_features2 on some platforms due to
689 	 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
690 	 * which made older superblock reading/writing routines swap it as a
691 	 * 64-bit value.
692 	 *
693 	 * For backwards compatibility, we make both slots equal.
694 	 *
695 	 * If we detect a mismatched field, we OR the set bits into the existing
696 	 * features2 field in case it has already been modified; we don't want
697 	 * to lose any features.  We then update the bad location with the ORed
698 	 * value so that older kernels will see any features2 flags. The
699 	 * superblock writeback code ensures the new sb_features2 is copied to
700 	 * sb_bad_features2 before it is logged or written to disk.
701 	 */
702 	if (xfs_sb_has_mismatched_features2(sbp)) {
703 		xfs_warn(mp, "correcting sb_features alignment problem");
704 		sbp->sb_features2 |= sbp->sb_bad_features2;
705 		mp->m_update_sb = true;
706 
707 		/*
708 		 * Re-check for ATTR2 in case it was found in bad_features2
709 		 * slot.
710 		 */
711 		if (xfs_sb_version_hasattr2(&mp->m_sb) &&
712 		   !(mp->m_flags & XFS_MOUNT_NOATTR2))
713 			mp->m_flags |= XFS_MOUNT_ATTR2;
714 	}
715 
716 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
717 	   (mp->m_flags & XFS_MOUNT_NOATTR2)) {
718 		xfs_sb_version_removeattr2(&mp->m_sb);
719 		mp->m_update_sb = true;
720 
721 		/* update sb_versionnum for the clearing of the morebits */
722 		if (!sbp->sb_features2)
723 			mp->m_update_sb = true;
724 	}
725 
726 	/* always use v2 inodes by default now */
727 	if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
728 		mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
729 		mp->m_update_sb = true;
730 	}
731 
732 	/*
733 	 * Check if sb_agblocks is aligned at stripe boundary
734 	 * If sb_agblocks is NOT aligned turn off m_dalign since
735 	 * allocator alignment is within an ag, therefore ag has
736 	 * to be aligned at stripe boundary.
737 	 */
738 	error = xfs_update_alignment(mp);
739 	if (error)
740 		goto out;
741 
742 	xfs_alloc_compute_maxlevels(mp);
743 	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
744 	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
745 	xfs_ialloc_compute_maxlevels(mp);
746 	xfs_rmapbt_compute_maxlevels(mp);
747 	xfs_refcountbt_compute_maxlevels(mp);
748 
749 	xfs_set_maxicount(mp);
750 
751 	/* enable fail_at_unmount as default */
752 	mp->m_fail_unmount = true;
753 
754 	error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
755 	if (error)
756 		goto out;
757 
758 	error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
759 			       &mp->m_kobj, "stats");
760 	if (error)
761 		goto out_remove_sysfs;
762 
763 	error = xfs_error_sysfs_init(mp);
764 	if (error)
765 		goto out_del_stats;
766 
767 	error = xfs_errortag_init(mp);
768 	if (error)
769 		goto out_remove_error_sysfs;
770 
771 	error = xfs_uuid_mount(mp);
772 	if (error)
773 		goto out_remove_errortag;
774 
775 	/*
776 	 * Set the minimum read and write sizes
777 	 */
778 	xfs_set_rw_sizes(mp);
779 
780 	/* set the low space thresholds for dynamic preallocation */
781 	xfs_set_low_space_thresholds(mp);
782 
783 	/*
784 	 * Set the inode cluster size.
785 	 * This may still be overridden by the file system
786 	 * block size if it is larger than the chosen cluster size.
787 	 *
788 	 * For v5 filesystems, scale the cluster size with the inode size to
789 	 * keep a constant ratio of inode per cluster buffer, but only if mkfs
790 	 * has set the inode alignment value appropriately for larger cluster
791 	 * sizes.
792 	 */
793 	mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
794 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
795 		int	new_size = mp->m_inode_cluster_size;
796 
797 		new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
798 		if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
799 			mp->m_inode_cluster_size = new_size;
800 	}
801 
802 	/*
803 	 * If enabled, sparse inode chunk alignment is expected to match the
804 	 * cluster size. Full inode chunk alignment must match the chunk size,
805 	 * but that is checked on sb read verification...
806 	 */
807 	if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
808 	    mp->m_sb.sb_spino_align !=
809 			XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) {
810 		xfs_warn(mp,
811 	"Sparse inode block alignment (%u) must match cluster size (%llu).",
812 			 mp->m_sb.sb_spino_align,
813 			 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size));
814 		error = -EINVAL;
815 		goto out_remove_uuid;
816 	}
817 
818 	/*
819 	 * Set inode alignment fields
820 	 */
821 	xfs_set_inoalignment(mp);
822 
823 	/*
824 	 * Check that the data (and log if separate) is an ok size.
825 	 */
826 	error = xfs_check_sizes(mp);
827 	if (error)
828 		goto out_remove_uuid;
829 
830 	/*
831 	 * Initialize realtime fields in the mount structure
832 	 */
833 	error = xfs_rtmount_init(mp);
834 	if (error) {
835 		xfs_warn(mp, "RT mount failed");
836 		goto out_remove_uuid;
837 	}
838 
839 	/*
840 	 *  Copies the low order bits of the timestamp and the randomly
841 	 *  set "sequence" number out of a UUID.
842 	 */
843 	mp->m_fixedfsid[0] =
844 		(get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
845 		 get_unaligned_be16(&sbp->sb_uuid.b[4]);
846 	mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
847 
848 	error = xfs_da_mount(mp);
849 	if (error) {
850 		xfs_warn(mp, "Failed dir/attr init: %d", error);
851 		goto out_remove_uuid;
852 	}
853 
854 	/*
855 	 * Initialize the precomputed transaction reservations values.
856 	 */
857 	xfs_trans_init(mp);
858 
859 	/*
860 	 * Allocate and initialize the per-ag data.
861 	 */
862 	error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
863 	if (error) {
864 		xfs_warn(mp, "Failed per-ag init: %d", error);
865 		goto out_free_dir;
866 	}
867 
868 	if (!sbp->sb_logblocks) {
869 		xfs_warn(mp, "no log defined");
870 		XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
871 		error = -EFSCORRUPTED;
872 		goto out_free_perag;
873 	}
874 
875 	/*
876 	 * Log's mount-time initialization. The first part of recovery can place
877 	 * some items on the AIL, to be handled when recovery is finished or
878 	 * cancelled.
879 	 */
880 	error = xfs_log_mount(mp, mp->m_logdev_targp,
881 			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
882 			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
883 	if (error) {
884 		xfs_warn(mp, "log mount failed");
885 		goto out_fail_wait;
886 	}
887 
888 	/* Make sure the summary counts are ok. */
889 	error = xfs_check_summary_counts(mp);
890 	if (error)
891 		goto out_log_dealloc;
892 
893 	/*
894 	 * Get and sanity-check the root inode.
895 	 * Save the pointer to it in the mount structure.
896 	 */
897 	error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
898 			 XFS_ILOCK_EXCL, &rip);
899 	if (error) {
900 		xfs_warn(mp,
901 			"Failed to read root inode 0x%llx, error %d",
902 			sbp->sb_rootino, -error);
903 		goto out_log_dealloc;
904 	}
905 
906 	ASSERT(rip != NULL);
907 
908 	if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) {
909 		xfs_warn(mp, "corrupted root inode %llu: not a directory",
910 			(unsigned long long)rip->i_ino);
911 		xfs_iunlock(rip, XFS_ILOCK_EXCL);
912 		XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
913 				 mp);
914 		error = -EFSCORRUPTED;
915 		goto out_rele_rip;
916 	}
917 	mp->m_rootip = rip;	/* save it */
918 
919 	xfs_iunlock(rip, XFS_ILOCK_EXCL);
920 
921 	/*
922 	 * Initialize realtime inode pointers in the mount structure
923 	 */
924 	error = xfs_rtmount_inodes(mp);
925 	if (error) {
926 		/*
927 		 * Free up the root inode.
928 		 */
929 		xfs_warn(mp, "failed to read RT inodes");
930 		goto out_rele_rip;
931 	}
932 
933 	/*
934 	 * If this is a read-only mount defer the superblock updates until
935 	 * the next remount into writeable mode.  Otherwise we would never
936 	 * perform the update e.g. for the root filesystem.
937 	 */
938 	if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
939 		error = xfs_sync_sb(mp, false);
940 		if (error) {
941 			xfs_warn(mp, "failed to write sb changes");
942 			goto out_rtunmount;
943 		}
944 	}
945 
946 	/*
947 	 * Initialise the XFS quota management subsystem for this mount
948 	 */
949 	if (XFS_IS_QUOTA_RUNNING(mp)) {
950 		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
951 		if (error)
952 			goto out_rtunmount;
953 	} else {
954 		ASSERT(!XFS_IS_QUOTA_ON(mp));
955 
956 		/*
957 		 * If a file system had quotas running earlier, but decided to
958 		 * mount without -o uquota/pquota/gquota options, revoke the
959 		 * quotachecked license.
960 		 */
961 		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
962 			xfs_notice(mp, "resetting quota flags");
963 			error = xfs_mount_reset_sbqflags(mp);
964 			if (error)
965 				goto out_rtunmount;
966 		}
967 	}
968 
969 	/*
970 	 * Finish recovering the file system.  This part needed to be delayed
971 	 * until after the root and real-time bitmap inodes were consistently
972 	 * read in.
973 	 */
974 	error = xfs_log_mount_finish(mp);
975 	if (error) {
976 		xfs_warn(mp, "log mount finish failed");
977 		goto out_rtunmount;
978 	}
979 
980 	/*
981 	 * Now the log is fully replayed, we can transition to full read-only
982 	 * mode for read-only mounts. This will sync all the metadata and clean
983 	 * the log so that the recovery we just performed does not have to be
984 	 * replayed again on the next mount.
985 	 *
986 	 * We use the same quiesce mechanism as the rw->ro remount, as they are
987 	 * semantically identical operations.
988 	 */
989 	if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
990 							XFS_MOUNT_RDONLY) {
991 		xfs_quiesce_attr(mp);
992 	}
993 
994 	/*
995 	 * Complete the quota initialisation, post-log-replay component.
996 	 */
997 	if (quotamount) {
998 		ASSERT(mp->m_qflags == 0);
999 		mp->m_qflags = quotaflags;
1000 
1001 		xfs_qm_mount_quotas(mp);
1002 	}
1003 
1004 	/*
1005 	 * Now we are mounted, reserve a small amount of unused space for
1006 	 * privileged transactions. This is needed so that transaction
1007 	 * space required for critical operations can dip into this pool
1008 	 * when at ENOSPC. This is needed for operations like create with
1009 	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1010 	 * are not allowed to use this reserved space.
1011 	 *
1012 	 * This may drive us straight to ENOSPC on mount, but that implies
1013 	 * we were already there on the last unmount. Warn if this occurs.
1014 	 */
1015 	if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1016 		resblks = xfs_default_resblks(mp);
1017 		error = xfs_reserve_blocks(mp, &resblks, NULL);
1018 		if (error)
1019 			xfs_warn(mp,
1020 	"Unable to allocate reserve blocks. Continuing without reserve pool.");
1021 
1022 		/* Recover any CoW blocks that never got remapped. */
1023 		error = xfs_reflink_recover_cow(mp);
1024 		if (error) {
1025 			xfs_err(mp,
1026 	"Error %d recovering leftover CoW allocations.", error);
1027 			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1028 			goto out_quota;
1029 		}
1030 
1031 		/* Reserve AG blocks for future btree expansion. */
1032 		error = xfs_fs_reserve_ag_blocks(mp);
1033 		if (error && error != -ENOSPC)
1034 			goto out_agresv;
1035 	}
1036 
1037 	return 0;
1038 
1039  out_agresv:
1040 	xfs_fs_unreserve_ag_blocks(mp);
1041  out_quota:
1042 	xfs_qm_unmount_quotas(mp);
1043  out_rtunmount:
1044 	xfs_rtunmount_inodes(mp);
1045  out_rele_rip:
1046 	xfs_irele(rip);
1047 	/* Clean out dquots that might be in memory after quotacheck. */
1048 	xfs_qm_unmount(mp);
1049 	/*
1050 	 * Cancel all delayed reclaim work and reclaim the inodes directly.
1051 	 * We have to do this /after/ rtunmount and qm_unmount because those
1052 	 * two will have scheduled delayed reclaim for the rt/quota inodes.
1053 	 *
1054 	 * This is slightly different from the unmountfs call sequence
1055 	 * because we could be tearing down a partially set up mount.  In
1056 	 * particular, if log_mount_finish fails we bail out without calling
1057 	 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1058 	 * quota inodes.
1059 	 */
1060 	cancel_delayed_work_sync(&mp->m_reclaim_work);
1061 	xfs_reclaim_inodes(mp, SYNC_WAIT);
1062  out_log_dealloc:
1063 	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1064 	xfs_log_mount_cancel(mp);
1065  out_fail_wait:
1066 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1067 		xfs_wait_buftarg(mp->m_logdev_targp);
1068 	xfs_wait_buftarg(mp->m_ddev_targp);
1069  out_free_perag:
1070 	xfs_free_perag(mp);
1071  out_free_dir:
1072 	xfs_da_unmount(mp);
1073  out_remove_uuid:
1074 	xfs_uuid_unmount(mp);
1075  out_remove_errortag:
1076 	xfs_errortag_del(mp);
1077  out_remove_error_sysfs:
1078 	xfs_error_sysfs_del(mp);
1079  out_del_stats:
1080 	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1081  out_remove_sysfs:
1082 	xfs_sysfs_del(&mp->m_kobj);
1083  out:
1084 	return error;
1085 }
1086 
1087 /*
1088  * This flushes out the inodes,dquots and the superblock, unmounts the
1089  * log and makes sure that incore structures are freed.
1090  */
1091 void
xfs_unmountfs(struct xfs_mount * mp)1092 xfs_unmountfs(
1093 	struct xfs_mount	*mp)
1094 {
1095 	uint64_t		resblks;
1096 	int			error;
1097 
1098 	xfs_icache_disable_reclaim(mp);
1099 	xfs_fs_unreserve_ag_blocks(mp);
1100 	xfs_qm_unmount_quotas(mp);
1101 	xfs_rtunmount_inodes(mp);
1102 	xfs_irele(mp->m_rootip);
1103 
1104 	/*
1105 	 * We can potentially deadlock here if we have an inode cluster
1106 	 * that has been freed has its buffer still pinned in memory because
1107 	 * the transaction is still sitting in a iclog. The stale inodes
1108 	 * on that buffer will have their flush locks held until the
1109 	 * transaction hits the disk and the callbacks run. the inode
1110 	 * flush takes the flush lock unconditionally and with nothing to
1111 	 * push out the iclog we will never get that unlocked. hence we
1112 	 * need to force the log first.
1113 	 */
1114 	xfs_log_force(mp, XFS_LOG_SYNC);
1115 
1116 	/*
1117 	 * Wait for all busy extents to be freed, including completion of
1118 	 * any discard operation.
1119 	 */
1120 	xfs_extent_busy_wait_all(mp);
1121 	flush_workqueue(xfs_discard_wq);
1122 
1123 	/*
1124 	 * We now need to tell the world we are unmounting. This will allow
1125 	 * us to detect that the filesystem is going away and we should error
1126 	 * out anything that we have been retrying in the background. This will
1127 	 * prevent neverending retries in AIL pushing from hanging the unmount.
1128 	 */
1129 	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1130 
1131 	/*
1132 	 * Flush all pending changes from the AIL.
1133 	 */
1134 	xfs_ail_push_all_sync(mp->m_ail);
1135 
1136 	/*
1137 	 * And reclaim all inodes.  At this point there should be no dirty
1138 	 * inodes and none should be pinned or locked, but use synchronous
1139 	 * reclaim just to be sure. We can stop background inode reclaim
1140 	 * here as well if it is still running.
1141 	 */
1142 	cancel_delayed_work_sync(&mp->m_reclaim_work);
1143 	xfs_reclaim_inodes(mp, SYNC_WAIT);
1144 
1145 	xfs_qm_unmount(mp);
1146 
1147 	/*
1148 	 * Unreserve any blocks we have so that when we unmount we don't account
1149 	 * the reserved free space as used. This is really only necessary for
1150 	 * lazy superblock counting because it trusts the incore superblock
1151 	 * counters to be absolutely correct on clean unmount.
1152 	 *
1153 	 * We don't bother correcting this elsewhere for lazy superblock
1154 	 * counting because on mount of an unclean filesystem we reconstruct the
1155 	 * correct counter value and this is irrelevant.
1156 	 *
1157 	 * For non-lazy counter filesystems, this doesn't matter at all because
1158 	 * we only every apply deltas to the superblock and hence the incore
1159 	 * value does not matter....
1160 	 */
1161 	resblks = 0;
1162 	error = xfs_reserve_blocks(mp, &resblks, NULL);
1163 	if (error)
1164 		xfs_warn(mp, "Unable to free reserved block pool. "
1165 				"Freespace may not be correct on next mount.");
1166 
1167 	error = xfs_log_sbcount(mp);
1168 	if (error)
1169 		xfs_warn(mp, "Unable to update superblock counters. "
1170 				"Freespace may not be correct on next mount.");
1171 
1172 
1173 	xfs_log_unmount(mp);
1174 	xfs_da_unmount(mp);
1175 	xfs_uuid_unmount(mp);
1176 
1177 #if defined(DEBUG)
1178 	xfs_errortag_clearall(mp);
1179 #endif
1180 	xfs_free_perag(mp);
1181 
1182 	xfs_errortag_del(mp);
1183 	xfs_error_sysfs_del(mp);
1184 	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1185 	xfs_sysfs_del(&mp->m_kobj);
1186 }
1187 
1188 /*
1189  * Determine whether modifications can proceed. The caller specifies the minimum
1190  * freeze level for which modifications should not be allowed. This allows
1191  * certain operations to proceed while the freeze sequence is in progress, if
1192  * necessary.
1193  */
1194 bool
xfs_fs_writable(struct xfs_mount * mp,int level)1195 xfs_fs_writable(
1196 	struct xfs_mount	*mp,
1197 	int			level)
1198 {
1199 	ASSERT(level > SB_UNFROZEN);
1200 	if ((mp->m_super->s_writers.frozen >= level) ||
1201 	    XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1202 		return false;
1203 
1204 	return true;
1205 }
1206 
1207 /*
1208  * xfs_log_sbcount
1209  *
1210  * Sync the superblock counters to disk.
1211  *
1212  * Note this code can be called during the process of freezing, so we use the
1213  * transaction allocator that does not block when the transaction subsystem is
1214  * in its frozen state.
1215  */
1216 int
xfs_log_sbcount(xfs_mount_t * mp)1217 xfs_log_sbcount(xfs_mount_t *mp)
1218 {
1219 	/* allow this to proceed during the freeze sequence... */
1220 	if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1221 		return 0;
1222 
1223 	/*
1224 	 * we don't need to do this if we are updating the superblock
1225 	 * counters on every modification.
1226 	 */
1227 	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1228 		return 0;
1229 
1230 	return xfs_sync_sb(mp, true);
1231 }
1232 
1233 /*
1234  * Deltas for the inode count are +/-64, hence we use a large batch size
1235  * of 128 so we don't need to take the counter lock on every update.
1236  */
1237 #define XFS_ICOUNT_BATCH	128
1238 int
xfs_mod_icount(struct xfs_mount * mp,int64_t delta)1239 xfs_mod_icount(
1240 	struct xfs_mount	*mp,
1241 	int64_t			delta)
1242 {
1243 	percpu_counter_add_batch(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1244 	if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
1245 		ASSERT(0);
1246 		percpu_counter_add(&mp->m_icount, -delta);
1247 		return -EINVAL;
1248 	}
1249 	return 0;
1250 }
1251 
1252 int
xfs_mod_ifree(struct xfs_mount * mp,int64_t delta)1253 xfs_mod_ifree(
1254 	struct xfs_mount	*mp,
1255 	int64_t			delta)
1256 {
1257 	percpu_counter_add(&mp->m_ifree, delta);
1258 	if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1259 		ASSERT(0);
1260 		percpu_counter_add(&mp->m_ifree, -delta);
1261 		return -EINVAL;
1262 	}
1263 	return 0;
1264 }
1265 
1266 /*
1267  * Deltas for the block count can vary from 1 to very large, but lock contention
1268  * only occurs on frequent small block count updates such as in the delayed
1269  * allocation path for buffered writes (page a time updates). Hence we set
1270  * a large batch count (1024) to minimise global counter updates except when
1271  * we get near to ENOSPC and we have to be very accurate with our updates.
1272  */
1273 #define XFS_FDBLOCKS_BATCH	1024
1274 int
xfs_mod_fdblocks(struct xfs_mount * mp,int64_t delta,bool rsvd)1275 xfs_mod_fdblocks(
1276 	struct xfs_mount	*mp,
1277 	int64_t			delta,
1278 	bool			rsvd)
1279 {
1280 	int64_t			lcounter;
1281 	long long		res_used;
1282 	s32			batch;
1283 
1284 	if (delta > 0) {
1285 		/*
1286 		 * If the reserve pool is depleted, put blocks back into it
1287 		 * first. Most of the time the pool is full.
1288 		 */
1289 		if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1290 			percpu_counter_add(&mp->m_fdblocks, delta);
1291 			return 0;
1292 		}
1293 
1294 		spin_lock(&mp->m_sb_lock);
1295 		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1296 
1297 		if (res_used > delta) {
1298 			mp->m_resblks_avail += delta;
1299 		} else {
1300 			delta -= res_used;
1301 			mp->m_resblks_avail = mp->m_resblks;
1302 			percpu_counter_add(&mp->m_fdblocks, delta);
1303 		}
1304 		spin_unlock(&mp->m_sb_lock);
1305 		return 0;
1306 	}
1307 
1308 	/*
1309 	 * Taking blocks away, need to be more accurate the closer we
1310 	 * are to zero.
1311 	 *
1312 	 * If the counter has a value of less than 2 * max batch size,
1313 	 * then make everything serialise as we are real close to
1314 	 * ENOSPC.
1315 	 */
1316 	if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1317 				     XFS_FDBLOCKS_BATCH) < 0)
1318 		batch = 1;
1319 	else
1320 		batch = XFS_FDBLOCKS_BATCH;
1321 
1322 	percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1323 	if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1324 				     XFS_FDBLOCKS_BATCH) >= 0) {
1325 		/* we had space! */
1326 		return 0;
1327 	}
1328 
1329 	/*
1330 	 * lock up the sb for dipping into reserves before releasing the space
1331 	 * that took us to ENOSPC.
1332 	 */
1333 	spin_lock(&mp->m_sb_lock);
1334 	percpu_counter_add(&mp->m_fdblocks, -delta);
1335 	if (!rsvd)
1336 		goto fdblocks_enospc;
1337 
1338 	lcounter = (long long)mp->m_resblks_avail + delta;
1339 	if (lcounter >= 0) {
1340 		mp->m_resblks_avail = lcounter;
1341 		spin_unlock(&mp->m_sb_lock);
1342 		return 0;
1343 	}
1344 	printk_once(KERN_WARNING
1345 		"Filesystem \"%s\": reserve blocks depleted! "
1346 		"Consider increasing reserve pool size.",
1347 		mp->m_fsname);
1348 fdblocks_enospc:
1349 	spin_unlock(&mp->m_sb_lock);
1350 	return -ENOSPC;
1351 }
1352 
1353 int
xfs_mod_frextents(struct xfs_mount * mp,int64_t delta)1354 xfs_mod_frextents(
1355 	struct xfs_mount	*mp,
1356 	int64_t			delta)
1357 {
1358 	int64_t			lcounter;
1359 	int			ret = 0;
1360 
1361 	spin_lock(&mp->m_sb_lock);
1362 	lcounter = mp->m_sb.sb_frextents + delta;
1363 	if (lcounter < 0)
1364 		ret = -ENOSPC;
1365 	else
1366 		mp->m_sb.sb_frextents = lcounter;
1367 	spin_unlock(&mp->m_sb_lock);
1368 	return ret;
1369 }
1370 
1371 /*
1372  * xfs_getsb() is called to obtain the buffer for the superblock.
1373  * The buffer is returned locked and read in from disk.
1374  * The buffer should be released with a call to xfs_brelse().
1375  *
1376  * If the flags parameter is BUF_TRYLOCK, then we'll only return
1377  * the superblock buffer if it can be locked without sleeping.
1378  * If it can't then we'll return NULL.
1379  */
1380 struct xfs_buf *
xfs_getsb(struct xfs_mount * mp,int flags)1381 xfs_getsb(
1382 	struct xfs_mount	*mp,
1383 	int			flags)
1384 {
1385 	struct xfs_buf		*bp = mp->m_sb_bp;
1386 
1387 	if (!xfs_buf_trylock(bp)) {
1388 		if (flags & XBF_TRYLOCK)
1389 			return NULL;
1390 		xfs_buf_lock(bp);
1391 	}
1392 
1393 	xfs_buf_hold(bp);
1394 	ASSERT(bp->b_flags & XBF_DONE);
1395 	return bp;
1396 }
1397 
1398 /*
1399  * Used to free the superblock along various error paths.
1400  */
1401 void
xfs_freesb(struct xfs_mount * mp)1402 xfs_freesb(
1403 	struct xfs_mount	*mp)
1404 {
1405 	struct xfs_buf		*bp = mp->m_sb_bp;
1406 
1407 	xfs_buf_lock(bp);
1408 	mp->m_sb_bp = NULL;
1409 	xfs_buf_relse(bp);
1410 }
1411 
1412 /*
1413  * If the underlying (data/log/rt) device is readonly, there are some
1414  * operations that cannot proceed.
1415  */
1416 int
xfs_dev_is_read_only(struct xfs_mount * mp,char * message)1417 xfs_dev_is_read_only(
1418 	struct xfs_mount	*mp,
1419 	char			*message)
1420 {
1421 	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1422 	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1423 	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1424 		xfs_notice(mp, "%s required on read-only device.", message);
1425 		xfs_notice(mp, "write access unavailable, cannot proceed.");
1426 		return -EROFS;
1427 	}
1428 	return 0;
1429 }
1430 
1431 /* Force the summary counters to be recalculated at next mount. */
1432 void
xfs_force_summary_recalc(struct xfs_mount * mp)1433 xfs_force_summary_recalc(
1434 	struct xfs_mount	*mp)
1435 {
1436 	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1437 		return;
1438 
1439 	spin_lock(&mp->m_sb_lock);
1440 	mp->m_flags |= XFS_MOUNT_BAD_SUMMARY;
1441 	spin_unlock(&mp->m_sb_lock);
1442 }
1443